Recognition algorithm on safe states of tower crane pins based on optimized Swin Transformer

doi:10.16265/j.cnki.issn1003-3033.2023.09.0173

Abstract

Abstract:

In order to reduce the hidden danger during tower crane operation, and improve the accuracy of machine vision when identifying states of pins, a recognition algorithm was proposed based on an optimized Swin Transformer. Firstly, A data set was created by collecting the pinned images of the tower crane on the construction site. Secondly, the safe state of pins was classified and encoded using the One-hot-coding method in the data set. Then, a recognition model for the safety status of pins was established based on an optimized Swin Transformer in which the loss function was adjusted. Through updating the gradient by the AdamW optimizer, a final model was obtained after 1000 training iterations. Finally, an experimental verification was conducted on the pinned image dataset. The results show that the proposed method can improve the identification ability of the safe state of tower crane pins, and its accuracy can reach 99.4%. The average accuracy, the average recall rate and the average specificity can reach 99.4%, 99.4%, and 99.6%, respectively. Its accuracy is higher than typical algorithms, such as DenseNet, ShuffleNet and EfficientNet. When opposed to the original Swin Transformer, the accuracy is also added by 3.6%.

Key words: Swin Transformer, tower crane, pins, safety state, state identification, data set

ZHOU Qinghui, LIU Haoshi. Recognition algorithm on safe states of tower crane pins based on optimized Swin Transformer[J]. China Safety Science Journal, 2023, 33(9): 63-68.

Figures/Tables 10

Fig.1

Fig.2

Tab.1

Tab.2

Tab.3

Tab.4

Evaluation index calculation formula

评价指标	公式
A	A= $T P + T N T P + T N + F P + F N$
P	P= $T P T P + F P$
R	R= $T P T P + F N$
S	S= $T N T N + F P$
AP	AP= $P 1 + P 2 + … + P N N$

Tab.4

Fig.3

Fig.4

Fig.5

Tab.5

References 17

[1]	赵挺生, 胡俊杰, 师玉栋, 等. 塔吊安装与拆卸作业安全性分析[J]. 中国安全科学学报, 2021, 31(10):32-38. doi: 10.16265/j.cnki.issn1003-3033.2021.10.005
	ZHAO Tingsheng, HU Junjie, SHI Yudong, et al. Safety analysis on tower crane installation and dismantling[J]. China Safety Science Journal, 2021, 31(10) :32-38. doi: 10.16265/j.cnki.issn1003-3033.2021.10.005
[2]	朱俊成, 郭泗宾, 张虎, 等. 塔机附着装置连接销轴的轴向位移与预防措施[J]. 建筑机械化, 2021, 43(1):65-67.
	ZHU Juncheng, GUO Sibin, ZHANG Hu, et al. Axial displacement of connecting pin of tower crane attachment device and preventive measures[J]. Construction Mechanization, 2021, 43(1):65-67.
[3]	GB/T 26471—2011,塔式起重机安装与拆卸规则[S].
	GB/T 26471-2011, Tower crane installation and disassembly rules[S].
[4]	杨静. 基于无人机视觉的起重机检测技术研究[J]. 电子设计工程, 2019, 27(21):72-75.
	YANG Jing. Application of unmanned aerial vehicle system in crane inspection[J]. Electronic Design Engineering, 2019, 27(21):72-75.
[5]	JULINA J K J, SHARMILA T S. Facial recognition using histogram of gradients and support vector machines[C]. 2017 International Conference on Computer, Communication and Signal Processing (ICCCSP), IEEE, 2017:1-5.
[6]	GRABNER M, GRABNER H, BISCHOF H. Fast approximated SIFT[C]. Computer Vision-ACCV 2006: 7^th Asian Conference on Computer Vision, 2006:918-927.
[7]	HE Lianghua, ZOU Cairong, ZHAO Li, et al. An enhanced LBP feature based on facial expression recognition[C]. 2005 IEEE Engineering in Medicine and Biology 27^th Annual Conference, IEEE, 2006:3300-3303.
[8]	SHIN Yeonjong. Effects of depth, width, and initialization: a convergence analysis of layer-wise training for deep linear neural networks[J]. Analysis and Applications, 2022, 20(1):73-119. doi: 10.1142/S0219530521500263
[9]	ZHANG Xiangyu, ZHOU Xinyu, LIN Mengxiao, et al. Shufflenet: an extremely efficient convolutional neural network for mobile devices[C]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018:6848-6856.
[10]	HUANG Gao, LIU Zhuang, MAATEN L, et al. Densely connected convolutional networks[C]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017:4700-4708.
[11]	TAN Mingxing, LE Quoc V. Efficientnet: rethinking model scaling for convolutional neural networks[C]. International Conference on Machine Learning, PMLR, 2019:6105-6114.
[12]	LIU Ze, LIN Yutong, CAO Yue, et al. Swin transformer: hierarchical vision transformer using shifted windows[C]. Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021:10 012-10 022.
[13]	LIN Tsungyi, GOYAL P, GIRSHICK R, et al. Focal loss for dense object detection[C]. Proceedings of the IEEE International Conference on Computer Vision, 2017:2980-2988.
[14]	赵欣然, 张琪, 王卫东, 等. 可燃性粉尘云的图像检测方法[J]. 中国安全科学学报, 2020, 30(4):8-13. doi: 10.16265/j.cnki.issn1003-3033.2020.04.002
	ZHAO Xinran, ZHANG Qi, WANG Weidong, et al. Image detection method of combustible dust cloud[J]. China Safety Science Journal, 2020, 30(4):8-13. doi: 10.16265/j.cnki.issn1003-3033.2020.04.002
[15]	LENG Zhaoqi, TAN Mingxing, LIU Chenxi, et al. PolyLoss: a polynomial expansion perspective of classification loss functions[C]. International Conference on Learning Representations, 202: DOI: 10.48550/arXiv.2204.12511.
[16]	FENG Lei, SHU Senlin, LIN Zhuoyi, et al. Can cross entropy loss be robust to label noise?[C]. Proceedings of the Twenty-Ninth International Conference on International Joint Conferences on Artificial Intelligence, 2021:2206-2212.
[17]	GONZALEZ S, MIIKKULAINEN R. Optimizing loss functions through multi-variate taylor polynomial parameterization[C]. Proceedings of the Genetic and Evolutionary Computation Conference, 2021:305-313.

销轴状态	总样本数	训练样本	测试样本
不安全状态	376	301	75
安全状态	733	587	146
较不安全状态	452	362	90

销轴状态	类别	编码
不安全状态	无开口销或者由铁丝替代	100
正常安全状态	正常的开口销	010
较不安全状态	开口销开口过小	001

混淆矩阵		真实值
混淆矩阵		真值	假值
预测值	真值	TP	FP
预测值	假值	FN	TN

评价指标	不安全状态	正常安全状态	较不安状态	A	AP
优化后	—	—	—	0.994	0.994
P	1.000	0.993	0.989	—	—
R	1.000	0.993	0.989	—	—
S	1.000	0.994	0.995	—	—
ShuffleNet	—	—	—	0.833	0.851
P	0.914	0.794	0.845	—	—
R	0.853	0.925	0.667	—	—
S	0.975	0.788	0.950	—	—
DenseNet	—	—	—	0.916	0.927
P	0.971	0.893	0.916	—	—
R	0.893	0.973	0.844	—	—
S	0.992	0.897	0.968	—	—
EfficientNet	—	—	—	0.936	0.937
P	0.960	0.932	0.920	—	—
R	0.960	0.945	0.900	—	—
S	0.987	0.939	0.968	—	—
Swin Transformer	—	—	—	0.958	0.961
P	1.000	0.959	0.924	—	—
R	0.973	0.959	0.944	—	—
S	1.000	0.964	0.968	—	—